Specifications
Product Family
PEGATRON AI server portfolio - GPU servers, rack-scale systems, GPU building blocks, AI storage and AI infrastructure switches
AS210-2T1-8H3
2U liquid-cooled server on NVIDIA HGX Rubin NVL8 with dual Intel Xeon 6 processors
AS210 Networking
NVIDIA Rubin GPUs with NVLink 6, ConnectX-9 SuperNICs, BlueField-4 DPUs and Spectrum-X Ethernet
MS303-2A1G-P60
GPU server supporting up to four liquid-cooled NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs with dual AMD EPYC 9005 processors
MS303 Memory
96 GB GDDR7 per RTX PRO 6000 Blackwell Server Edition GPU
RA4800-64H3 and RA4800-72H3
Rack-scale liquid-cooled architectures for ultra-high-density AI deployment
Rack-Scale Scope
Announced at GTC 2026 alongside NVIDIA Vera Rubin NVL72 platforms
Older Generation Coverage
NVIDIA HGX Blackwell series and AMD Instinct MI355X rack solutions also offered
Server CPUs
Support for 6th Gen AMD EPYC server processors across the general-purpose range
Networking Products
400G and 800G AI infrastructure switches for edge-to-core AI data centre connectivity
NIC Integration
AMD Pensando Pollara 400 AI NIC validated with 6th Gen AMD EPYC platforms
Form Factors
2U GPU servers, rack-scale integrated racks, multi-node platforms and AI storage systems
Cooling
Liquid cooling used across the Rubin NVL8 and RTX PRO 6000 platforms
Workloads
Generative AI training and inference, data processing, simulation and visualisation
Memory Subsystem
High-bandwidth GDDR7 on RTX PRO Blackwell cards; HBM with NVLink 6 interconnect on Rubin platforms
Interconnect
NVLink 6 GPU-to-GPU plus Spectrum-X Ethernet for scale-out
Integrated Rack Options
RA4800 rack series for factory-integrated, liquid-cooled deployments
Regional Presence
Portfolio promoted across Japan IT Week, CloudFest Europe and COMPUTEX 2026
Target Deployments
AI factories, enterprise AI data centres, HPC clusters and neocloud build-outs
Cross-Brand
Supermicro SYS-821GE-TNHR, Wiwynn GB200 NVL72, QuantaGrid D75E, Aivres KR series, Ingrasys GB300 NVL72
Overview
PEGATRON builds AI servers across the whole ladder rather than a single tier. At the top, the AS210-2T1-8H3 is a 2U liquid-cooled server on NVIDIA HGX Rubin NVL8 with dual Intel Xeon 6 processors, using NVLink 6 for GPU-to-GPU traffic and ConnectX-9 SuperNICs, BlueField-4 DPUs and Spectrum-X Ethernet for scale-out. The RA4800-64H3 and RA4800-72H3 extend the same architecture to factory-integrated liquid-cooled racks.
Below the rack scale, the MS303-2A1G-P60 targets enterprise multi-workload deployments: up to four liquid-cooled NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, each with 96 GB of GDDR7, paired with dual AMD EPYC 9005 processors. It suits organisations that need broad GPU capability - generative AI, data processing, simulation and visualisation - without moving to a full NVL rack.
PEGATRON also supplies the surrounding infrastructure rather than only the compute sled: 400G and 800G AI infrastructure switches, validated AI NICs, multi-node platforms and AI storage systems. That breadth matters in large deployments, where the switch fabric and storage tier determine whether the GPUs stay fed.
Key Benefits
Full ladder from 2U HGX Rubin NVL8 systems to liquid-cooled rack-scale platforms, RTX PRO Blackwell density with 96 GB GDDR7 per GPU for enterprise multi-workload use, and matched networking and storage products from the same vendor to simplify fabric and AI storage integration.
Applications
AI factory and data centre build-outs, large-scale generative AI training and inference, HPC clusters, enterprise multi-workload AI, visualisation and simulation, and AI storage tiers feeding GPU clusters at 400G and 800G.
Request a Quote — PEGATRON AI SERVERS — HGX RUBIN NVL8, VERA RUBIN NVL72 AND RTX PRO PLATFORMS
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →